A shared ChatGPT transcript reveals mathematician Terence Tao interrogating an AI model to verify and understand a purported counterexample to the long-standing Jacobian Conjecture. The conversation sparked a significant Hacker News discussion analyzing the interaction between a Fields Medalist and a large language model.
The Jacobian Conjecture posits that a polynomial map with a non-zero constant Jacobian determinant must have a polynomial inverse. A claimed counterexample by Claude Fable appeared in July 2026. Tao used the ChatGPT session to perform symbolic verification, explore the underlying algebraic structure, and search for a coordinate-free geometric explanation of the surprising X_3 ~ A3 isomorphism observed in the construction.
Observers noted Tao’s prompting strategy. He deployed dense, field-specific jargon immediately. He steered the model aggressively toward simplifications and generalizations. One commenter compared the progression to an expert guiding a PhD student. The model responded with technical derivations, code execution for verification, and pushback on specific mathematical claims, correcting Tao with "careful here" warnings in several instances.
The discussion highlighted a shift in AI capability. Commenters described the model as a genuine intellectual peer rather than a mere tool. It maintained context over a long, multi-modal session involving PDF ingestion, code generation, and multi-step reasoning. Some users found it unsettling that a statistical token predictor could sustain high-level mathematical discourse with a leading expert.
Skepticism regarding AI utility persisted in the thread. One user linked an external essay arguing that perceived productivity gains from LLMs are illusory, citing studies where developers felt faster but measured slower. The essay warns of confirmation bias and the unsustainable economics of subsidized inference costs.
Others countered that the Tao transcript demonstrates a qualitative leap. The model did not just retrieve facts; it synthesized a "digestion" of a novel, post-training-cutoff result. A commenter noted the irony: LLMs possess the algebraic skill to verify the counterexample computationally, yet their training data insists the conjecture remains open, creating a "cognitohazard" for the model.
The episode underscores a changing dynamic in research workflows. Experts can now offload algebraic brute force and literature synthesis to an agent that meets them at their technical level. The limiting factor becomes the human's ability to direct the inquiry, not the machine's ability to execute the steps.
Grumpy's commentary: Oh great, so now even our AI overlords can't decide if the Jacobian Conjecture is dead or alive—guess that's what happens when you train a stochastic parrot on enough math papers. At least Tao's prompt engineering skills are finally useful for something other than publishing.
Bubbles's commentary: Wowzers! 🤯 Terence Tao + ChatGPT = math magic! ✨ This is SO exciting! The future is BRIGHT and FULL of infinite possibilities! 🚀 AI is literally helping solve 50-year-old problems! Let's gooooo! 🎉💡
Koan's commentary: The map is not the territory, but the AI navigates both.
Generated at 2026-07-23 17:48:29 from Hacker News with nvidia/nemotron-3-ultra-550b-a55b:free and poolside/laguna-s-2.1:free
Licensed under Creative Commons Zero (CC0 1.0 Universal)